MiniMax-H3 AI 视频 API
MiniMax-H3 可通过 Flatkey 异步 /v1/videos 流程使用。打开控制台前可配置 768P 或 2K 分辨率、4–15 秒、支持的比例和 AIGC 水印字段。
输出
预览使用 Seedance 2.0 生成
请求摘要
- 接口
- POST /v1/videos
- 模型 ID
- MiniMax-H3
- 画面比例
- 9:16
- 分辨率
- 768P
- 时长
- 6s
- 参考媒体
- 无
仅供预览。登录后即可向 Flatkey 提交此请求。
{
"model": "MiniMax-H3",
"content": [
{
"type": "text",
"text": "Creative direction: A teal reusable bottle with water droplets stands on bright marble beside an open canvas tote. One hand grips the bottle, lifts it and slides it into the tote side pocket. End with the bottle still visible above the pocket and daylight catching its wet surface.\n\nVisual treatment: Photoreal lifestyle commercial with soft window daylight, teal enamel and tactile canvas and marble textures.\n\nShot sequence: 1. Frame the bottle and the open tote together; 2. Follow the hand as it lifts the bottle into the side pocket; 3. Hold the packed tote with the bottle visible and upright\n\nContinuity: Keep the product silhouette, cap or hinge, materials and prop positions stable; hold the final hero frame.\nExclude: extra people or props, text, logos and watermarks."
}
],
"resolution": "768P",
"duration": 6,
"ratio": "9:16",
"aigc_watermark": false
}能力
MiniMax-H3 API 性能与可用性
当 Flatkey 实时请求流量足够时,这里会显示 MiniMax-H3 的请求遥测;不据此推断基准测试或质量评分。
API
MiniMax-H3 API 使用与请求活动
此区块反映 Flatkey 对 MiniMax-H3 的实时请求;收集到足够流量前不会报告数据。
价格
MiniMax-H3 价格
下方价格来自当前模型的 Flatkey 定价数据,以及定价 API 返回的可见分组。
充值余额
图像、视频、音频和文本模型共用同一个 Flatkey 余额和 API Key。
- 模型类型
- 文生视频
- API
- /v1/videos
- 计费依据
- 秒
- 时长
- 4–15s
- 画面比例
- 21:9 · 16:9 · 4:3 · 1:1 · 3:4 · 9:16 · adaptive
MiniMax-H3 视频生成与制作能力
下方只展示已记录的集成字段,不推断基准测试或质量承诺。
文生视频与图生视频
通过 /v1/videos 从文字脚本或已设计的画面开始生成场景。
参考素材驱动镜头
使用参考素材保持产品、角色或分镜在镜头发展过程中的视觉连贯性。
可控的动作与节奏
先确定镜头、主体动作、时长和构图如何推进,再得到可用于剪辑的片段。
音频感知的制作衔接
明确处理可选声音和输出设置,让测试片段顺利进入剪辑与交付流程。
对比
MiniMax-H3 ComfyUI 与本地设置:已核实边界
下方只展示已记录的集成字段,不推断基准测试或质量承诺。
| 能力 | 未核实的本地假设 | 已核实的托管字段 |
|---|---|---|
| 分辨率 | 768P 基础值;2K 可通过托管重新生成 | 768P 或 2K |
| 时长 | 4–15 秒 | 4–15 秒 |
| 比例 | 文生视频使用固定比例;图像或参考素材路线使用 adaptive | 文生视频使用固定比例;图像或参考素材路线使用 adaptive |
| AIGC 水印 | 未核实 | 明确的布尔值 |
| ComfyUI 或本地权重 | 未核实 | Flatkey 目录未核实 |
| 上游与 Flatkey | 开源/本地使用遵循 MiniMax 上游文档 | Flatkey 提供路由并为托管 API 访问计费 |
接入前先选对请求方式
核对支持的工作流、当前配置的实时费用,以及 Flatkey 实际使用的参数契约。
当目录提供固定单价时,根据当前 Playground 配置和实时目录价格计算。
- 当前配置
- 768P · 6s
- 实时单价
- from $0.080 / 秒 · 4s min
- 本次预估
- $0.480 $0.080 × 6s
最终费用以实际提交的请求和账户分组为准。Token 或分层计费无法简化成一个固定预估值。
工作流与 Model ID
只使用标记为可用的工作流。Flatkey 对外保持一个模型 ID,由请求体选择输入方式。
文生视频
通过文本提示词和支持的输出参数生成视频。
- 模型 ID
MiniMax-H3- 接口
/v1/videos
图生视频
让一张或多张参考图动起来,并尽量保持主体与构图。
- 模型 ID
MiniMax-H3- 接口
/v1/videos
参考生视频
使用已文档化的图片、视频或音频参考素材指导结果。
- 模型 ID
MiniMax-H3- 接口
/v1/videos
参数兼容矩阵
以下取值与 Playground 和请求预览使用同一份模型配置。
resolution分辨率768P · 2K
默认值: 768Pduration时长4–15
默认值: 6ratio画面比例21:9 · 16:9 · 4:3 · 1:1 · 3:4 · 9:16 · adaptive
默认值: 16:9aigc_watermarkAIGC 水印开 · 关
默认值: 关image参考素材上限9 个文件
接口: /v1/videosvideo参考素材上限3 个文件
接口: /v1/videosaudio参考素材上限3 个文件
接口: /v1/videos从首次测试到生产接入
Playground 与 API 使用同一套请求契约,并明确处理参数校验、余额、任务状态和产物获取。
请求生命周期
- 1
校验模型、提示词、参考素材和输出设置。
- 2
提交 POST 请求并保存返回的任务 ID。
- 3
轮询任务状态或处理已配置的回调。
- 4
从产物接口获取文件,并在失效前保存到自己的存储。
生产检查项
提交前拦截不支持的尺寸、时长、比例、格式或参考素材数量。
分别处理无效密钥、余额不足、限流和账户分组价格。
明确指出被拒绝的提示词或素材,让用户只替换失败的输入。
保留任务 ID,限制重试次数,且不能把超时当成生成成功。
提示词库
MiniMax-H3 提示词指南:产品、UGC 与分镜
从具体主体、动作、镜头或构图和明确的输出设置开始;请求字段与提示词分开配置。
使用 Seedance 2.0 生成
Creative direction: A person in a yellow raincoat waits beside a bench at a rainy station, notices a torn letter, and picks it up to read. Move from the wet platform to a close view of the paper and hold on the reader's concerned expression. Visual treatment: Rainy character drama with subdued station colors and a yellow raincoat focal accent. Shot sequence: 1. Establish the raincoat figure beside the station bench.; 2. Move closer as the torn letter is noticed and lifted.; 3. Hold the concerned face with the letter visible and rain continuing behind. Continuity: Keep the character's face, clothing and key prop consistent; preserve screen direction across cuts. Exclude: extra people or props, text, logos and watermarks.
使用 Seedance 2.0 生成
Creative direction: A teal reusable bottle with water droplets stands on bright marble beside an open canvas tote. One hand grips the bottle, lifts it and slides it into the tote side pocket. End with the bottle still visible above the pocket and daylight catching its wet surface. Visual treatment: Photoreal lifestyle commercial with soft window daylight, teal enamel and tactile canvas and marble textures. Shot sequence: 1. Frame the bottle and the open tote together; 2. Follow the hand as it lifts the bottle into the side pocket; 3. Hold the packed tote with the bottle visible and upright Continuity: Keep the product silhouette, cap or hinge, materials and prop positions stable; hold the final hero frame. Exclude: extra people or props, text, logos and watermarks.
使用 Seedance 2.0 生成
Creative direction: A man walks from a dark interior into an empty glass-domed observatory at dawn. Follow him toward the telescope as cool sky light fills the dome, then settle behind the figure and instrument facing the bright horizon. Visual treatment: Live-action observatory reveal with cool dawn skylight and a dark foreground silhouette. Shot sequence: 1. Follow the man out of the dark interior into the glass dome.; 2. Continue toward the telescope as dawn sky fills the background.; 3. Settle behind the figure and telescope facing the bright horizon. Continuity: Maintain scene geography, subject scale and motivated light direction throughout the camera move. Exclude: extra people or props, text, logos and watermarks.
使用 Seedance 2.0 生成
Creative direction: A cyberpunk fighter crouches on a wet neon plaza, bursts into a sprint, and slides to a stop. A blue circular energy shield opens in front of the fighter as the low camera tracks the movement and holds on the glowing barrier. Visual treatment: Rendered cyberpunk action with neon reflections and localized blue shield emission. Shot sequence: 1. Start low on the crouched fighter in the wet plaza.; 2. Track the acceleration and slide, keeping foot contact legible.; 3. Stop with the fighter as the blue shield opens in front. Continuity: Maintain character design, equipment geometry and travel direction from first to final frame. Exclude: extra people or props, text, logos and watermarks.
使用 Seedance 2.0 生成
Creative direction: Miniature island and map pieces assemble on a blue tabletop around a white lighthouse. The land sections lock together, the lighthouse rises upright, and its lamp turns on while the overhead view reveals the completed coastal model. Visual treatment: Miniature map animation with distinct island layers and a blue tabletop. Shot sequence: 1. Establish separated island pieces and the lighthouse from overhead.; 2. Slide the land sections together and raise the lighthouse upright.; 3. Turn on the lamp and hold the completed coastal model. Continuity: Keep demonstration props, layout and hand contact consistent so the action remains legible. Exclude: extra people or props, text, logos and watermarks.
使用 Seedance 2.0 生成
Creative direction: A chrome ring-and-ribbon sculpture rotates above a reflective black floor under blue and red light. The metal loops bend and morph into a new interlocking shape while the camera holds a symmetrical gallery-style composition. Visual treatment: Reflective abstract sculpture animation with chrome, black and controlled blue-red light. Shot sequence: 1. Center the chrome loops above the black reflective floor.; 2. Rotate the sculpture while ribbons bend into the next interlocking shape.; 3. Slow the rotation and hold the new silhouette with its reflection aligned. Continuity: Keep the main subject, palette and light behavior consistent as the visual pattern develops. Exclude: extra people or props, text, logos and watermarks.
为什么选择 Flatkey
为什么通过 Flatkey 使用 MiniMax-H3?
在从测试请求走向生产流量时,清楚展示 MiniMax-H3 的事实、请求字段和价格边界。
集中展示已记录控制项
选择 768P 或 2K、4–15 秒、支持的比例,并明确设置 AIGC 水印布尔值。
附带提示词指南
产品、UGC 和分镜示例会说明主体、动作、镜头与结尾,不重复空泛的电影感提示词。
异步结果路径
保存 POST /v1/videos 返回的 task ID,任务完成后获取生成内容。
不做未经核实的承诺
目录未核实 Flatkey 原生 ComfyUI 交付或本地权重;本页不作此暗示。
API
MiniMax-H3 视频 API 设置
保存异步请求返回的 task ID,准备好后再获取结果。
curl https://router.flatkey.ai/v1/videos \ -H "Content-Type: application/json" \ -H "Authorization: Bearer sk-***" \ -d '{"model":"MiniMax-H3","content":[{"type":"text","text":"Creative direction: A person in a yellow raincoat waits beside a bench at a rainy station, notices a torn letter, and picks it up to read. Move from the wet platform to a close view of the paper and hold on the reader'"'"'s concerned expression.\n\nVisual treatment: Rainy character drama with subdued station colors and a yellow raincoat focal accent.\n\nShot sequence: 1. Establish the raincoat figure beside the station bench.; 2. Move closer as the torn letter is noticed and lifted.; 3. Hold the concerned face with the letter visible and rain continuing behind.\n\nContinuity: Keep the character'"'"'s face, clothing and key prop consistent; preserve screen direction across cuts.\nExclude: extra people or props, text, logos and watermarks."}],"resolution":"768P","duration":6,"ratio":"16:9","aigc_watermark":false}' # Poll for the result (replace TASK_ID with the returned task id) curl "https://router.flatkey.ai/v1/videos/TASK_ID" \ -H "Authorization: Bearer sk-***"
常见问题
MiniMax H3
视频 API: 价格与 ComfyUI 问题
价格、兼容性、限制与模型专属请求细节。